How Distributed Is Visual Category Information in Human Occipito-Temporal Cortex? An fMRI Study  Mona Spiridon, Nancy Kanwisher  Neuron  Volume 35, Issue 6, Pages 1157-1165 (September 2002) DOI: 10.1016/S0896-6273(02)00877-2

Figure 1 Calculation of the Percentage of Correct Discrimination for a Given Pair of Stimulus Categories A and B Each circle represents an activation map for a given category based on half the data collected for that category and subject. Thus A1 is the activation map for category A based on half the data and A2 is the activation map based on the other half. r represents the correlation coefficient between two activation maps. The percent correct discrimination between categories A and B is 100% if all of the following inequalities are correct: rA1A2 > rA1B2, rA1A2 > rA2B1, rB1B2 > rA2B1, rB1B2 > rA1B2. If only two or three of these relations are correct, the percent correct discrimination drops to 75% or 50%, respectively. The percent correct discrimination for all seven categories is the average of the percent correct discrimination over 84 pairwise comparisons. Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)

Figure 2 Examples of Stimuli for the Category “Chairs” There are eight stimuli each in groups 1 to 4 and 32 stimuli in group 5. The other categories are handled in the same fashion. Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)

Figure 3 Mean Percentage Pairwise Correct Discrimination and Standard Error across Six Subjects for Different Discrimination Types Based on All Visually Active Voxels The two sets of activation maps used in the analysis were produced from stimulus sets that were either identical grayscale photographs (blue), photographs with different viewpoints (green), photographs with different exemplars (red), or different formats (i.e., photographs and line drawings) (yellow). Chance is 50%. The percent correct discrimination is calculated from the average across: (A) all pairwise comparisons between the seven object categories, (B) all pairwise comparisons between two small man-made objects, (C) all pairwise comparisons between faces and small man-made objects, and (D) all pairwise comparisons between houses and small man-made objects. Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)

Figure 4 Mean Percentage of Correct Discrimination across Six Subjects as a Function of the Number of Voxels in Different Cortical Regions For each subject and each subset size, the percentage of correct discrimination is averaged over many different subsets of randomly selected voxels. Chance level is 50%. Overall performance on (A) all 84 discriminations for the seven categories, (B) face versus object discrimination averaged over four pairwise comparisons (faces versus chairs, faces versus bottles, faces versus scissors, faces versus shoes), (C) same as (B) but for houses instead of faces, (D) object discrimination averaged over six pairwise comparisons (chairs versus bottles, chairs versus scissors, chairs versus shoes, bottles versus scissors, bottle versus shoes, scissors versus shoes). Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)

Figure 5 Mean Percentage of Correct Discrimination and Standard Deviation in Three Different Regions The performance is calculated for different types of pairwise discriminations in the entire FFA, the entire PPA, and a 30 voxel cluster in the calcarine sulcus (probably V1), based on activations from identical images. A t test comparing performance to chance level (50%) produces the following p values: FFA: faces versus objects (p < 0.001), houses versus objects (p = 0.08), objects versus objects (p = 0.14); PPA: faces versus objects (p = 0.28), houses versus objects (p = 0.003), objects versus objects (p = 0.08); V1: faces versus objects (p < 0.001), houses versus objects (p = 0.02), objects versus objects (p = 0.08). The stars (*) indicate p values below 0.05. Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)

Figure 6 Mean across Subjects of the Number of Voxels Selective for Faces, Houses, Chairs, Bottles, and Shoes The selective voxels are the voxels for which the activation is significantly higher for one category than for the other ones (p < 10−5, uncorrected for multiple comparisons). Neuron 2002 35, 1157-1165DOI: (10.1016/S0896-6273(02)00877-2)